A geometric q -character formula for snake modules
aa r X i v : . [ m a t h . QA ] M a y A GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES BING DUAN AND RALF SCHIFFLER
Abstract.
Let C be the category of finite dimensional modules over the quantum affine algebra U q ( b g ) of a simple complex Lie algebra g . Let C − be the subcategory introduced by Hernandezand Leclerc. We prove the geometric q -character formula conjectured by Hernandez and Leclercin types A and B for a class of simple modules called snake modules introduced by Mukhin andYoung. Moreover, we give a combinatorial formula for the F -polynomial of the generic kernelassociated to the snake module. As an application, we show that snake modules correspondto cluster monomials with square free denominators and we show that snake modules are realmodules. We also show that the cluster algebras of the category C are factorial for Dynkin types A , D , E . Introduction
Let g be a simple complex Lie algebra and let U q ( b g ) be the corresponding quantum affine algebrawith quantum parameter q ∈ C × not a root of unity. Denote by C the category of finite dimensional U q ( b g )-modules. The simple modules in C have been classified in [3, 4] by Chari and Pressley interms of Drinfeld polynomials. In [13], Frenkel and Reshetikhin attached a q -character to everymodule in C and showed that the simple modules are determined up to isomorphism by their q -characters. Moreover, the simple modules are parametrized by the highest dominant monomials intheir q -characters.Cluster algebras were introduced in [14] by Fomin and Zelevinsky as a tool for studying canon-ical bases in Lie theory. A cluster algebra is a commutative algebra with a distinguished set ofgenerators, the cluster variables . These cluster variables are constructed by a recursive methodcalled mutation, which is determined by the choice of a quiver Q without loops and 2-cycles. Givena cluster algebra A ( Q ), every cluster variable can be expressed as a Laurent polynomial with in-teger coefficients with respect to any given cluster [14] and this Laurent polynomial has positivecoefficients [23]. A cluster monomial is a product of cluster variables from the same cluster. It wasproved in [7] that the set of all cluster monomials is linearly independent.1.1. Category C . A connection between representations of quantum affine algebras and clusteralgebras was discovered by Hernandez and Leclerc in [18], where a monoidal categorification ofcertain cluster algebras was given. One significant aspect of a monoidal categorification is that, if
Mathematics Subject Classification.
Key words and phrases. cluster algebra; quantum affine algebra; snake module; geometric character formula.The first author was supported by China Scholarship Council as a Joint PHD student to visit Department ofMathematics at the University of Connecticut and he would like to thank Department of Mathematics for hospitalityduring his visit. He was also partially supported by the National Natural Science Foundation of China (Grant No.11771191).The second author was supported by NSF CAREER Grant DMS-1254567, NSF Grant DMS-1800860 and by theUniversity of Connecticut. it exists, it implies the positivity of the cluster variables and the linear independence of the clustermonomials, see [18, Proposition 2.2].The monoidal categorification is realized as a subcategory of the category C as follows. Let I bethe vertex set of the Dynkin diagram of g for Dynkin types A , D , E and I = I ∪ I be a partitionof I such that every edge connects a vertex of I with a vertex of I . Let C ℓ , ℓ ≥
0, be the fullsubcategory of C whose objects V satisfy the following property. For any composition factor S of V and every i ∈ I , the roots of the i -th Drinfeld polynomial of S belong to { q − k − ξ i | ≤ k ≤ ℓ } ,where ξ i = 1 if i ∈ I and ξ i = 0 if i ∈ I .For Dynkin types A and D , it has been shown in [18] that the category C is a monoidalcategorification of a cluster algebra of the same Dynkin type. This result was extended to Dynkintypes A , D , E by Nakajima in [27], see also [19]. In [31], Qin proved that every cluster monomialcorresponds to a simple module in C for Dynkin types A , D , E .A simple module M in C is said to be real if M ⊗ M is simple [21], and M is said to be prime if it cannot be written as a non-trivial tensor product of modules [6].1.2. Category C − . In [20], Hernandez and Leclerc considered a much larger subcategory C − of C which contains, up to spectral shifts, all the simple finite-dimensional U q ( b g )-modules. Theyshowed that the Grothendieck ring of C − has a cluster algebra structure [20, Theorem 5.1], andthey proposed two conjectures. Conjecture 1.1. [20, Conjecture 5.2]
The cluster monomials of the cluster algebra are in bijectionwith the isomorphism classes of real simple objects in C − . The second conjecture uses the theory of quivers with potentials developed in [8,9]. In [20, Section5.2.2], Hernandez and Leclerc associated to every simple U q ( b g )-module M a so-called generic kernel K ( M ), which is a module over the Jacobian algebra of the quiver with potential. They showedthat, up to normalization, the truncated q -character of a Kirillov-Reshetikhin module is equal to the F -polynomial of the associated generic kernel, and they conjectured the following generalization. Conjecture 1.2. [20, Conjecture 5.3]
Up to normalization, the truncated q -character of a realsimple module in C − is equal to the F -polynomial of the associated generic kernel. Snake modules.
In this paper, we prove both conjectures in Dynkin types A and B for snakemodules , a class of simple U q ( b g )-modules introduced by Mukhin and Young in [24, 25]. In [24], theyintroduced a purely combinatorial method to compute q -characters for snake modules of types A and B , and in [25], they used snake modules to construct extended T -systems for types A and B .In [10], it was shown that all prime snake modules are real and that they correspond to some clustervariables in the cluster algebra constructed by Hernandez and Leclerc.Our first main theorem is the following. Theorem 1.3. (Theorem 3.2 and Remark 3.4) Let L ( m ) be a prime snake module in C − . Then upto normalization, the truncated q -character of L ( m ) is equal to the F -polynomial of the associatedgeneric kernel K ( m ) . More precisely, χ − q ( L ( m )) = mF K ( m ) . Replacing the module K ( m ) by a direct sum, we obtain a similar geometric character formula forarbitrary snake module of types A and B . This proves Conjecture 1.2 for snake modules and gives a geometric algorithm for the truncated q -characters. As a slight generalization of Theorem 3.4 of [10], we show in Theorem 4.2 that snakemodules are real modules. GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 3 Then in Theorem 3.9 and Remark 3.10 (3), we give a combinatorial formula for the F -polynomialof the generic kernel K ( m ) associated to a snake module L ( m ) as a sum over certain non-overlappingpaths in a subset of the Z × Z -grid determined by m . This result uses the model of Mukhin andYoung [24, 25]. As a consequence, we also obtain a combinatorial method to find the dimensionvector of K ( m ) as well as all its submodules. Furthermore we show that K ( m ) is always rigid andit is indecomposable if the snake module L ( m ) is prime.As an application, we prove Conjecture 1.1 for snake modules in the first part of the followingtheorem. Theorem 1.4. (Theorems 4.1 and 4.4) The truncated q -character of a snake module L ( m ) isa cluster monomial. Moreover the denominator of this cluster monomial is square free and isparametrized by the support of K ( m ) as a representation of the quiver with potential. It is natural to ask whether all cluster variables with square free denominators correspond tosnake modules. This is not the case. However, the only counter examples we found are moduleswhose truncated q -characters are not equal to their ordinary q -characters.Lastly, the study of square free denominators led us to questions of factorization in the clusteralgebra. Applying the results of [11], we include a proof that the C cluster algebras of Hernandez-Leclerc are factorial for Dynkin types A , D , E .This paper is organized as follows. In Section 2, we briefly review basic materials on clusteralgebras, quantum affine algebras, snake modules, and Hernandez and Leclerc’s results. Section 3is on our geometric character formula for snake modules and Section 4 is devoted to the study ofdenominator vectors of cluster monomials corresponding to snake modules. In the last section, weprove that the C cluster algebras are factorial for Dynkin types A , D , E .2. Preliminaries
Quivers and cluster algebras.
We recall the definition of the cluster algebras introducedin [20]. Let C = ( c ij ) i,j ∈ I be an indecomposable n × n Cartan matrix of finte type. Then thereexists a diagonal matrix D = diag( d i | i ∈ I ) with positive entries such that B = DC = ( b ij ) i,j ∈ I is symmetric. Let t = max { d i | i ∈ I } . Thus t = C is of type A n , D n , E , E , E , C is of type B n , C n or F , C is of type G . Let e G be the infinite quiver with vertex set e G = I × Z and arrows ( i, r ) → ( j, s ) if b ij = 0and s = r + b ij . It needs to be pointed out that e G has two isomorphic connected components, seeLemma 2.2 of [20]. We pick one of the two components and denote it by G with vertex set G . Weconsider the full subquiver G − of G with vertex set G − = G ∩ ( I × Z ≤ ), see Figure 1.Let z = { z i,r | ( i, r ) ∈ G − } and let A be the cluster algebra defined by the initial seed ( z , G − ).The cluster algebra A is a cluster algebra of infinite rank.Let Y − = { Y ± i,r | ( i, r ) ∈ G − } be a new set of indeterminates over Q . For ( i, r ) ∈ G − , we define k i,r to be the unique positive integer k satisfying0 < kb ii − | r | ≤ b ii . In other words, ( i, r ) is the k th vertex in its column, counting from the top. BING DUAN AND RALF SCHIFFLER (2 , , −
1) (3 , − , − , −
3) (3 , − , − , −
5) (3 , − ... ... ... z z ttttt $ $ ❏❏❏❏❏ $ $ ❏❏❏❏❏ z z ttttt O O z z ttttt $ $ ❏❏❏❏❏ O O $ $ ❏❏❏❏❏ O O z z ttttt O O z z ttttt $ $ ❏❏❏❏❏ O O O O O O O O O O (cid:31) (cid:31) ❄❄❄❄❄❄❄❄❄ (cid:127) (cid:127) ⑧⑧⑧⑧⑧⑧⑧⑧⑧ (2 , , −
2) (1 , − , −
3) (2 , − , −
6) (1 , − , −
7) (2 , − , −
10) (1 , − , −
11) (2 , − ... ... ... O O O O O O O O O O O O O O O O O O O O O O O O O O z z ttttt $ $ ❏❏❏❏❏ z z ttttt $ $ ❏❏❏❏❏ z z ttttt (cid:31) (cid:31) ❄❄❄❄❄❄❄❄❄ $ $ ❏❏❏❏❏ z z ttttt $ $ ❏❏❏❏❏ z z ttttt $ $ ❏❏❏❏❏ z z ttttt (cid:31) (cid:31) ❄❄❄❄❄❄❄❄❄ Figure 1.
The quvier G − in type A (left) and the quvier G − in type B (right).For ( i, r ) ∈ G − , we perform the substitution z i,r = k i,r − Y j =0 Y i,r + jb ii . (2.1)Note that z i,r z i,r + b ii = Y i,r for ( i, r + b ii ) ∈ G − .Let Γ be the same quiver as G but with vertex set Γ = { ( i, r − d i ) : ( i, r ) ∈ G } . Let Γ − be thefull subquiver of Γ with vertex set Γ − = Γ ∩ ( I × Z ≤ ), see Figure 2.In this paper, we let g be of type A or B . We work in the full subcategory C − of C whoseobjects have all their composition factors of the form L ( m ), where m is a monomial in the variables Y i,r ∈ Y − .2.2. Quantum affine algebras.
Let g be a simple complex Lie algebra whose Dynkin diagramhas vertex set I and h ∨ be the dual Coxeter number of g , see Table 1. Let b g be the correspondinguntwisted affine Lie algebra which is realized as a central extension of the loop algebra g ⊗ C [ t, t − ].Let U q ( b g ) be the Drinfeld-Jimbo quantum enveloping algebra (quantum affine algebra for short) of b g with parameter q ∈ C ∗ not a root of unity, see [4]. GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 5 (2 , − , −
2) (3 , − , − , −
4) (3 , − , − , −
6) (3 , − ... ... ... z z ttttt $ $ ❏❏❏❏❏ $ $ ❏❏❏❏❏ z z ttttt O O z z ttttt $ $ ❏❏❏❏❏ O O $ $ ❏❏❏❏❏ O O z z ttttt O O z z ttttt $ $ ❏❏❏❏❏ O O O O O O O O O O (cid:31) (cid:31) ❄❄❄❄❄❄❄❄❄ (cid:127) (cid:127) ⑧⑧⑧⑧⑧⑧⑧⑧⑧ (2 , − , −
3) (1 , − , −
5) (2 , − , −
7) (1 , − , −
9) (2 , − , −
11) (1 , − , −
13) (2 , − ... ... ... O O O O O O O O O O O O O O O O O O O O O O O O O O z z ttttt $ $ ❏❏❏❏❏ z z ttttt $ $ ❏❏❏❏❏ z z ttttt (cid:31) (cid:31) ❄❄❄❄❄❄❄❄❄ $ $ ❏❏❏❏❏ z z ttttt $ $ ❏❏❏❏❏ z z ttttt $ $ ❏❏❏❏❏ z z ttttt (cid:31) (cid:31) ❄❄❄❄❄❄❄❄❄ Figure 2.
The quvier Γ − in type A (left) and the quvier Γ − in type B (right). g A n B n C n D n E E E F G t h ∨ n + 1 2 n − n + 1 2 n − Table 1.
Dual Coxeter numbersLet U q ( g ) be the quantum enveloping algebra. Recall that a U q ( g )-module V is of type 1 if it is adirect sum of its weight subspaces. A U q ( b g )-module V is said to be of type 1 if the central element c / acts as the identity on V , and if V is of type 1 as a module U q ( g ). Let C be the category offinite-dimensional U q ( b g )-modules of type 1. Every finite-dimensional simple U q ( b g )-module can beobtained from a type 1 module by twisting with an automorphism of U q ( b g ), see [4, 5].Let K ( C ) be the Grothendieck ring of C . Let P be the free abelian multiplicative group ofmonomials in infinitely many formal variables ( Y i,a ) i ∈ I ; a ∈ C × . The q -character of an object M in C is defined as an injective ring homomorphism χ q from K ( C ) to the ring Z P = Z [ Y ± i,a ] i ∈ I ; a ∈ C × ofLaurent polynomial in infinitely many formal variables.In this paper, we will be concerned only with polynomials involving the subset of variables Y i,aq r , a ∈ C × , ( i, r ) ∈ G . For simplicity of notation, we write Y i,r for Y i,aq r .A monomial in Z P is called dominant (respectively, anti-dominant ) if it does not contain a factor Y − i,r (respectively, Y i,r ) with ( i, r ) ∈ G . Following [13], for ( i, r ) ∈ Γ , define v i,r := A − i,r = Y − i,r − d i Y − i,r + d i Y j : c ji = − Y j,r Y j : c ji = − Y j,r − Y j,r +1 Y j : c ji = − Y j,r − Y j,r Y j,r +2 , (2.2) BING DUAN AND RALF SCHIFFLER where the c ij are the entries of the Cartan matrix. It follows that A i,r is a Laurent monomial inthe variables Y j,s with ( j, s ) ∈ G , see Section 2.3.2 of [19].For any simple object V in C , it was shown by Frenkel and Mukhin [12] that the q -charactercan be expressed as χ q ( V ) = m + (1 + X p M p ) , where m + ∈ Z P is a monomial in the variables Y i,r , ( i, r ) ∈ G , with positive powers, hence m + isa dominant monomial, and each M p is a product of factors A − i,r , ( i, r ) ∈ Γ . The monomial m + iscalled the highest weight monomial of V . There is a partial order ≤ on P defined by m ≤ m ′ if and only if m ′ m − is a monomial generated by A i,r , ( i, r ) ∈ Γ . Then m + is maximum with respect to ≤ .Every simple object in C can be parametrized by the highest weight monomial occurring in its q -character [3, 13]. The highest weight monomial is dominant, but in general the highest weightmonomial is not the only dominant monomial occurring in q -characters. Given a dominant mono-mial m , one can construct the corresponding simple module L ( m ).A simple module L ( m ) is called special or minuscule if m is the only dominant monomial oc-curring in χ q (( L ( m )), see Definition 10.1 of [26] or Section 5.2.2 of [18]. It is anti-special if there isexactly one anti-dominant monomial occurring in its q -character. Clearly, a special or anti-specialmodule must be simple. A simple module is called thin if any weight space of the simple modulehas no dimension greater than 1.Following [18], define the truncated q -character χ − q ( L ( m )) to be the Laurent polynomial obtainedfrom χ q ( L ( m )) by deleting all the monomials involving variables Y i,r Y − . In other words, χ − q ( L ( m )) ∈ Z [ Y ± i,r | ( i, r ) ∈ G − ]. By Proposition 3.10 of [20], χ − q is an injective ring homomorphismfrom the Grothendieck ring of C − to Z [ Y ± i,r | ( i, r ) ∈ G − ].2.3. Paths.
Define a subset
X ⊂ I × Z and an injective mapping ι : X → Z × Z as follows.Type A n : Let X := { ( i, k ) ∈ I × Z : i − k ≡ } and ι ( i, k ) = ( i, k ) . Type B n : Let X := { ( n, k ) : k ∈ Z } ⊔ { ( i, k ) ∈ I × Z : i < n and k ≡ } and ι ( i, k ) = (2 i, k ) , if i < n and 2 n + k − i ≡ , (4 n − − i, k ) , if i < n and 2 n + k − i ≡ , (2 n − , k ) , if i = n. Following [24, 25], for every ( i, k ) ∈ X , a set P i,k of paths is defined as follows. Here a path is afinite sequence of points in the plane R . We write ( j, ℓ ) ∈ p if ( j, ℓ ) is a point of the path p . In ourdiagrams, we connect consecutive points of a path by line segments for illustrative purposes only.The following is the case of type A n . For all ( i, k ) ∈ I × Z , let P i,k = { ((0 , y ) , (1 , y ) , . . . , ( n + 1 , y n +1 )) : y = i + k,y n +1 = n + 1 − i + k, and y j +1 − y j ∈ { , − } , ≤ j ≤ n } . In other words, a path in P i,k must start at (0 , i + k ) and end at ( n + 1 , n + 1 − i + k ) and eachstep between them can either go up one unit or go down one unit. So | P i,k | = (cid:0) n +1 i (cid:1) .Note that the cardinality of P i,k is equal to the number of Young tableaux that fit in an i × ( n + 1 − i ) rectangle. GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 7 Figure 3.
In type A : illustration of the paths in P , .The sets C ± p of upper and lower corners of a path p = (( r, y r )) ≤ r ≤ n +1 ∈ P i,k are defined asfollows (see Figure 3): C + p = { ( r, y r ) ∈ p : r ∈ I, y r − = y r + 1 = y r +1 } ,C − p = { ( r, y r ) ∈ p : r ∈ I, y r − = y r − y r +1 } . The following is the case of type B n . Fix an ε such that 0 < ε < /
2, for all ℓ ∈ Z , the set P n,ℓ is defined as follows.For all ℓ ≡ P n,ℓ = { ((0 , y ) , (2 , y ) , . . . , (2 n − , y n − ) , (2 n − , y n − ) , (2 n − , y n )) : y = ℓ + 2 n − ,y i +1 − y i ∈ { , − } , ≤ i ≤ n − , and y n − y n − ∈ { ǫ, − − ǫ }} . For all ℓ ≡ P n,ℓ = { ((4 n − , y ) , (4 n − , y ) , . . . , (2 n + 2 , y n − ) , (2 n, y n − ) , (2 n − , y n )) : y = ℓ + 2 n − ,y i +1 − y i ∈ { , − } , ≤ i ≤ n − , and y n − y n − ∈ { ǫ, − − ǫ }} . In other words, a path in P n,ℓ must start at (0 , ℓ + 2 n −
1) or at (4 n − , ℓ + 2 n −
1) and then itcan either go up or go down until (2 n − , y n ). So | P n,ℓ | = 2 n .For all ( i, k ) ∈ X , i < n , let P i,k = { ( a , a , . . . , a n , a n , . . . , a , a ) : ( a , a , . . . , a n ) ∈ P n,k − (2 n − i − , ( a , a , . . . , a n ) ∈ P n,k +(2 n − i − , and a n − a n = (0 , y ) where y > } . In other words, a path in P i,k , i < n , must start at (0 , k + 2 i ) (respectively, (4 n − , k + 2 i )) and endat (4 n − , k + 4 n − i −
2) (respectively, (0 , k + 4 n − i − | P n,ℓ | > (cid:0) n − i (cid:1) .The sets C ± p of upper and lower corners of a path p = (( j r , ℓ r )) ≤ r ≤| p |− ∈ P i,k , where | p | is thenumber of points in the path p , are defined as follows: C + p = ι − { ( j r , ℓ r ) ∈ p : j r
6∈ { , n − , n − } , ℓ r − > ℓ r , ℓ r +1 > ℓ r }⊔ { ( n, ℓ ) ∈ X : (2 n − , ℓ − ǫ ) ∈ p and (2 n − , ℓ + ǫ ) p } ,C − p = ι − { ( j r , ℓ r ) ∈ p : j r
6∈ { , n − , n − } , ℓ r − < ℓ r , ℓ r +1 < ℓ r }⊔ { ( n, ℓ ) ∈ X : (2 n − , ℓ − ǫ ) p and (2 n − , ℓ + ǫ ) ∈ p } . These definitions are illustrated in Figures 4 and 5.In order to subsequently describe our F -polynomials, it is helpful to define the notion of cell . Wecall a region in P i,k a cell if it is a minimal region enclosed by paths. For every cell, we define thecoordinate of the cell as follows. If the cell is a square or a square missing a corner, its coordinate BING DUAN AND RALF SCHIFFLER
Figure 4.
In type B : left, P , ; right, P , .1 2 3 2 1024681012 1 2 3 2 1024681012 Figure 5.
In type B : left, P , ; right, P , .is defined as the coordinate of the intersection of two diagonals. If the cell is a right triangle, itscoordinate is defined as the coordinate of the midpoint of its hypotenuse. It is obvious that for any( i, k ) ∈ G , our cell coordinate is an element of Γ .We also need the following notations in this paper. For all ( i, k ) ∈ X , let p + i,k be the highestpath which is the unique path in P i,k with no lower corners and p − i,k the lowest path which is theunique path in P i,k with no upper corners. Let p, p ′ be paths. We say that p is strictly above p ′ or GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 9 p ′ is strictly below p if ( x, y ) ∈ p and ( x, z ) ∈ p ′ = ⇒ y < z. Snake modules.
A simple module L ( m ) is called a Kirillov-Reshetikhin module if m is of theform m = k − Y j =0 Y i,r + jb ii , ( i ∈ I, r ∈ Z , k ≥ , (2.3)and is usually denoted by W ( i ) k,r .For completeness we recall the definition of snake module introduced by Mukhin and Youngin [24, 25]. Let ( i, k ) ∈ X . A point ( i ′ , k ′ ) ∈ X is said to be in snake position with respect to ( i, k )if Type A n : k ′ − k ≥ | i ′ − i | + 2 and k ′ − k ≡ | i ′ − i | (mod 2) . Type B n : i = i ′ = n : k ′ − k ≥ k ′ − k ≡ ,i = i ′ = n or i ′ = i = n : k ′ − k ≥ | i ′ − i | + 3 and k ′ − k ≡ | i ′ − i | − ,i < n and i ′ < n : k ′ − k ≥ | i ′ − i | + 4 and k ′ − k ≡ | i ′ − i | (mod 4) . The point ( i ′ , k ′ ) is in minimal snake position to ( i, k ) if k ′ − k is equal to the given lower bound.The point ( i ′ , k ′ ) is in prime snake position to ( i, k ) if Type A n : min { n + 2 − i − i ′ , i + i ′ } ≥ k ′ − k ≥ | i ′ − i | + 2 and k ′ − k ≡ | i ′ − i | (mod 2) . Type B n : i = i ′ = n : 4 n − ≥ k ′ − k ≥ k ′ − k ≡ ,i = i ′ = n or i ′ = i = n : 2 i ′ + 2 i − ≥ k ′ − k ≥ | i ′ − i | + 3 and k ′ − k ≡ | i ′ − i | − ,i < n and i ′ < n : 2 i ′ + 2 i ≥ k ′ − k ≥ | i ′ − i | + 4 and k ′ − k ≡ | i ′ − i | (mod 4) . A finite sequence ( i t , k t ), 1 ≤ t ≤ T , T ∈ Z ≥ , of points in X is called a snake (respectively, prime snake , minimal snake ) if for all 2 ≤ t ≤ T , the point ( i t , k t ) is in snake position (respectively,prime snake position, minimal snake position) with respect to ( i t − , k t − ) [24, 25].The simple module L ( m ) is called a snake module (respectively, prime snake module , mini-mal snake module ) if m = Q Tt =1 Y i t ,k t for some snake (respectively, prime snake, minimal snake)( i t , k t ) ≤ t ≤ T [24, 25]. In this case, we say that ( i t , k t ) ≤ t ≤ T is the snake of L ( m ). Theorem 2.1. [10, Section 4.1]
The snake modules of type A n or B n are precisely the U q ( b g ) -modules L ( m ) with highest weight monomial m = N Y j =1 k j − Y s =0 Y i j ,r + b ijij s + P j − ℓ =1 n ℓ , (2.4) where r ∈ Z , i j ∈ I, k j ≥ for ≤ j ≤ N ; furthermore n ℓ = b i ℓ i ℓ ( k ℓ −
1) + 2 t + t | i ℓ +1 − i ℓ | + 2 tj ℓ + ε i ℓ ,i ℓ +1 , where j ℓ ∈ Z ≥ for ≤ ℓ ≤ N − , ε i,j = ( g is of type A n , − δ in − δ jn g is of type B n , t = ( g is of type A n , g is of type B n , where δ ij is the Kronecker delta, and we use the convention P ℓ =1 n ℓ = 0 . In particular L ( m ) is aprime snake module if j ℓ satisfies the following bounded conditions:Type A n : 0 ≤ j ℓ ≤ ( n − max { i ℓ , i ℓ +1 } if i ℓ + i ℓ +1 ≥ n + 1;min { i ℓ , i ℓ +1 } − if i ℓ + i ℓ +1 < n + 1 . Type B n : 0 ≤ j ℓ ≤ min { i ℓ , i ℓ +1 } − . Example 2.2. (1)
Every Kirillov-Reshetikhin module is a snake module, by taking N = 1 inTheorem 2.1. (2) Minimal affinizations introduced in [2] (see also [32] ) are snake modules, by taking i i > . . . > i N and j ℓ = 0 in Theorem 2.1. (3) For examples of snake modules which are not Kirillov-Reshetikhin modules or minimalaffinizations see Examples 3.5–3.7.
A graphic interpretation is that in G − , the upper bound of the integer j ℓ is the minimumdistance from the columns with vertex labelings ( i ℓ , − ) and ( i ℓ +1 , − ) to the leftmost column andthe rightmost column respectively. The value of j ℓ is associated to the multiplicity of the left andright operators defined in Section 5.5 of [10].Let ( i ℓ , k ℓ ) ∈ X , 1 ≤ ℓ ≤ T , be a snake of length T ∈ Z ≥ and p ℓ ∈ P i ℓ ,k ℓ . We say that a T -tuple of paths ( p , . . . , p T ) is non-overlapping if p s is strictly above p t for all 1 ≤ s < t ≤ T . Let P ( i t ,k t ) ≤ t ≤ T = { ( p , . . . , p T ) : p t ∈ P i t ,k t , ≤ t ≤ T, ( p , . . . , p T ) is non-overlapping } . Mukhin and Young have proved the following theorem.
Theorem 2.3 ([24, Theorem 6.1]; [25, Theorem 6.5]) . Let ( i ℓ , k ℓ ) ∈ X , ≤ ℓ ≤ T , be a snake oflength T ∈ Z ≥ . Then χ q ( L ( T Y ℓ =1 Y i ℓ ,k ℓ )) = X ( p ,...,p T ) ∈ P ( iℓ,kℓ )1 ≤ ℓ ≤ T T Y ℓ =1 m ( p ℓ ) , (2.5) where the mapping m is defined by m : G ( i,k ) ∈X P i,k −→ Z [ Y ± j,ℓ ] ( j,ℓ ) ∈X p m ( p ) = Y ( j,ℓ ) ∈ C + p Y j,ℓ Y ( j,ℓ ) ∈ C − p Y − j,ℓ . Moreover, the module L ( Q Tℓ =1 Y i ℓ ,k ℓ ) is thin, special and anti-special. Remark 2.4.
A snake module L ( Q Tℓ =1 Y i ℓ ,k ℓ ) is prime if and only if for all ≤ t ≤ T the paths p + i t ,k t ∈ P i t ,k t and p − i t − ,k t − ∈ P i t − ,k t − are overlapping. In view of Theorem 2.3, the q -characters of snake modules of types A n and B n with length T aregiven by a set of T -tuples of non-overlapping paths, the path in each T -tuple is non-overlapping.This property is called the non-overlapping property .For any two paths p , p ∈ P i,k , p can be obtained from p by a sequence of moves, seeLemma 5.8 of [24]. We say that p ≤ p if m ( p ) ≤ m ( p ).In the following we list some known facts about snake modules. GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 11 Theorem 2.5. [25, Proposition 3.1]
A snake module is prime if and only if its snake is prime.Every snake module can be uniquely written as a tensor product of prime snake modules (up topermutation).
Theorem 2.6. [10, Theorem 3.4,Theorem 5.9]
Prime snake modules are real and they correspondto some cluster variables in the cluster algebra A . Moreover, in Theorem 4.1 of [25], Mukhin and Young introduced a set of 3-term recurrence rela-tions satisfied by q -characters of prime snake modules, called extended T -system, which generalizesthe usual T -system. Moreover, in Theorem 4.1 of [10], the authors introduced a system of equa-tions satisfied by q -characters of prime snake modules, called S -system, which contains the usual T -system. In fact the equations in the S -system can be interpreted as cluster transformations inthe cluster algebra A where the initial cluster variables correspond to certain Kirillov-Reshetikhinmodules.2.5. Quivers with potentials.
Following [20], for every i = j with c ij = 0, and every ( i, r ) ∈ Γ − ,we have in Γ − an oriented cycle with length 2 + | c ij | : ( i, r ) | | ①①①①①①①①①①①①①①①①①① ( j, r + b ij ) " " ❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊ ( i, r + 2 b ij + b ii ) O O ✤✤✤✤✤✤ ( i, r + 2 b ij ) O O A potential S is defined as the formal (infinite) sum for all these oriented cycles up to cyclicpermutations, see Section 3 of [8]. Hence in Γ − , all the cyclic derivatives of S , introduced inDefinition 3.1 of [8], are finite sums of paths. Indeed, a given arrow of Γ − can only occur in a finitenumber of summands.Let R be the set of all cyclic derivatives of S . Let J be the two-sided ideal of the path algebra C Γ − generated by R . Following [9, 20], one defines the Jacobian algebra A = C Γ − /J . Then A isan infinite-dimensional C -algebra.Let M be a finite-dimensional A -module, and e ∈ N Γ − be a dimension vector. Let Gr e ( M ) bethe quiver Grassmannian of M . Thus Gr e ( M ) is the variety of submodules of M with dimensionvector e . This is a projective complex variety. Denote by χ (Gr e ( M )) its Euler characteristic.Following [9, 20], define the F -polynomial of M as a polynomial in the indeterminates v i,r , ( i, r ) ∈ Γ − , as follows: F M = X e ∈ N Γ − χ (Gr e ( M )) Y ( i,r ) ∈ Γ − v e i,r i,r . It was shown in [9] that for any finite-dimensional M , F M is a monic polynomial with constantterm equal to 1. Following Section 4.5.2 of [20], let ℓ ∈ Z < and let Γ − ℓ be the full subquiver of Γ − with vertexset (Γ − ) ℓ := { ( i, m ) ∈ Γ − | m ≥ ℓ } . Let S ℓ be the sum of all cycles in the potential S which only involve vertices of (Γ − ) ℓ , called atruncation of S . Let J ℓ be the two-sided ideal of C Γ − ℓ generated by all cyclic derivatives of S ℓ andlet A ℓ = C Γ − ℓ /J ℓ be the truncated Jacobian algebra at height ℓ . Denote by π : C Γ − ℓ → A ℓ the natural projection.It has been shown in Proposition 4.17 of [20] that for any ℓ , A ℓ is finite-dimensional and thequiver with potential (Γ − ℓ , J ℓ ) is rigid, namely, every cycle is cyclically equivalent to an element of J ℓ .2.6. q -characters and F -polynomials. Let m be a dominant monomial in the variables Y i,r for( i, r ) ∈ G − . Following [15, 20], for each ( i, r ) ∈ G − , define b y i,r = Y ( i,r ) → ( j,s ) z j,s Y ( j,s ) → ( i,r ) z − j,s . It was shown in Lemma 4.15 of [20] that b y i,r = A − i,r − d i for ( i, r ) ∈ G − , so b y i,r is a monomial in thevariables Y i,s , ( i, s ) ∈ G − by (2.2).Using [15, Corollary 6.3], Hernandez and Leclerc gave the following formula for a cluster variablein terms of its F -polynomial and g -vector. Every cluster variable x of A has the following form x = z g x F x ( b y ) . (2.6)On the other hand, in [12], the truncated q -character χ − q ( L ( m )) is expressed as χ − q ( L ( m )) = mP m , (2.7)where P m is a polynomial with integer coefficients in the variables { A − i,r − d i | ( i, r ) ∈ G − } and hasconstant term 1. Thus, by [20], if L ( m ) is a cluster variable of A , then m = z g ( m ) , where theinteger vector g ( m ) ∈ Z G − is the g -vector of L ( m ).Let I i,r be the indecomposable injective A -module at vertex ( i, r ) ∈ Γ − . Motivated by quiverswith potentials [9] and cluster character [28,29], Hernandez and Leclerc defined the following notionof generic kernel. Definition 2.7. [20, Definition 4.5 and Section 5.2.2]
Let K ( m ) be the kernel of a generic A -modulehomomorphism from the injective A -module I ( m ) − to the injective A -module I ( m ) + , where I ( m ) + = M g i,r ( m ) > I ⊕ g i,r ( m ) i,r − d i , I ( m ) − = M g i,r ( m ) < I ⊕| g i,r ( m ) | i,r − d i . The support of K ( m ) is the collection of all points ( j, s ) ∈ Γ − such that the ( j, s )-component of K ( m ) is nonzero. We denote by Supp( K ( m )) the support of K ( m ).In [20], Hernandez and Leclerc proposed the following conjecture. Conjecture 2.8. [20, Conjecture 5.3]
Let L ( m ) be a real simple U q ( b g ) -mdoule in C − . Then upto normalization, the truncated q -character of L ( m ) is equal to the F -polynomial of the associatedgeneric kernel. More precisely, χ − q ( L ( m )) = mF K ( m ) , GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 13 where the variables v i,r of the F -polynomial are evaluated as in (2.2). In Theorem 4.8 of [20], Hernandez and Leclerc proved Conjecture 2.8 for Kirillov-Reshetikhinmodules, that is, up to renormalizing, the truncated q -character of the Kirillov-Reshetikhin module W ( i ) k,r − d i (2 k − is equal to the F -polynomial of the generic kernel F K ( i ) k,r , where K ( i ) k,r is the kernel ofa generic A -module homomorphism from I i,r to I i,r − kb ii . We will prove Conjecture 2.8 for snakemodules in Theorem 3.2.2.7. A formula for the lowest weight monomial.
Recall that t = max { d i | i ∈ I } as definedin Section 2.1. As a generalization of Remark 4.14 of [20], we can calculate the dimension vectorsof the A -module K ( i ) k,r for r ≤ d i (2 k − − th ∨ . Indeed, by Lemma 6.8 and Corollary 6.9 of [12],the lowest monomial of χ q ( Q ks =1 Y i,r − d i (2 s − ) is equal to Q ks =1 Y − v ( i ) ,r − d i (2 s − th ∨ , where v is theinvolution of I defined by w ( α i ) = − α v ( i ) , where w is the longest element in the Weyl group of g .Using Theorem 4.8 of [20], we can calculate the lowest monomial, which corresponds to the termin the F -polynomial for the trivial submodule K ( i ) k,r ⊂ K ( i ) k,r . Thus k Y s =1 Y − v ( i ) ,r − d i (2 s − th ∨ = k Y s =1 Y i,r − d i (2 s − ! Y ( j,s ) ∈ Γ − v d j,s ( K ( i ) k,r ) j,s , where ( d j,s ( K ( i ) k,r )) ( j,s ) ∈ N Γ − is the dimension vector of K ( i ) k,r .In the next section, we will introduce a combinatorial method to calculate the dimension vectorof the A -module K ( m ) associated to the snake module L ( m ).3. A geometric character formula for snake modules
In this section, we show that the geometric character formula conjectured by Hernandez andLeclerc holds for snake modules of types A and B . We give a combinatorial formula for the F -polynomial of the generic kernel K ( m ) associated to the snake module L ( m ). As a consequence,we obtain a combinatorial method to compute the dimension vector of K ( m ) as well as all itssubmodules.3.1. A geometric character formula for snake modules.
We first give a description of the g -vector g ( m ) := ( g i,s ) ( i,s ) ∈ G − for arbitrary prime snake module L ( m ). Proposition 3.1.
Let L ( m ) be a prime snake module with highest weight monomial m of theform (2.4). Then we can rewrite m = z g ( m ) := Y ( i,s ) ∈ G − z g i,s ( m ) i,s , where g i,s ( m ) = if ( i, s ) = ( i j , r + P j − ℓ =1 n ℓ ) and r + P j − ℓ =1 n ℓ ≤ , − if ( i, s ) = ( i j , r + P j − ℓ =1 n ℓ + b i j i j k j ) and r + P j − ℓ =1 n ℓ + b i j i j k j ≤ , otherwise . Here j = 1 , · · · , N as in (2.4). Proof.
From Theorem 2.1, it follows that every prime snake module L ( m ) is a U q ( b g )-module withhighest weight monomial m of the form (2.4). Thus m is a product of terms of the form k j − Y s =0 Y i j ,r + P j − ℓ =1 n ℓ + b ijij s . Because of (2.3), for any 1 ≤ j ≤ N , L k j − Y s =0 Y i j ,r + P j − ℓ =1 n ℓ + b ijij s is a Kirillov-Reshitikhin module. Now the result follows from Theorem 2.6 and Proposition 4.16of [20]. (cid:3) We are now ready for the main result of this section. The following theorem gives a positiveanswer to the Hernandez-Leclerc Conjecture (Conjecture 2.8) for snake modules.
Theorem 3.2.
Let L ( m ) be a prime snake module in C − . Then up to normalization, the truncated q -character of L ( m ) is equal to the F -polynomial of the associated generic kernel K ( m ) . Moreprecisely, χ − q ( L ( m )) = mF K ( m ) , where F K ( m ) is a polynomial in the variables (2.2).Proof. Recall that A is the cluster algebra defined in Section 2.1. We use the characterization of m from Theorem 2.1. The fact that L ( m ) ∈ C − implies that each index of Y in the formula (2.4)is a vertex in G − . This implies that for some integer N ,( i N , r + N − X ℓ =1 n ℓ + b i N i N ( k N − ∈ G − . (3.1)In particular, the second coordinate of (3.1) is non-positive. Thus r + P j − ℓ =1 n ℓ + b i j i j ( k j − ≤ j = 1 , · · · , N .By Theorem 2.6, the truncated q -character χ − q ( L ( m )) is a cluster variable x of A . By Proposi-tion 3.1, the g -vector of x is given by g i,s ( m ) = i, s ) = ( i j , r + P j − ℓ =1 n ℓ ) , − i, s ) = ( i j , r + P j − ℓ =1 n ℓ + b i j i j k j ) and r + P j − ℓ =1 n ℓ + b i j i j k j ≤ , , where we use that r + P j − ℓ =1 n ℓ ≤
0, because of (3.1).For ℓ <
0, let ( G − ) ℓ := { ( i, r + d i ) : ( i, r ) ∈ (Γ − ) ℓ } and z − ℓ = { z i,r | ( i, r ) ∈ ( G − ) ℓ } . We denoteby G − ℓ the same quiver as Γ − ℓ , but with vertices labeled by ( G − ) ℓ . Clearly, the cluster variable x is a Laurent polynomial in the variables of z − ℓ for some ℓ ≪
0, and can be regarded as a clustervariable of the cluster algebra A ℓ defined by the initial seed ( z − ℓ , G − ℓ ).The rest of the proof is similar to the proof of Theorem 4.8 in [20]. Since the quiver with potential(Γ − ℓ , J ℓ ) is rigid, we can apply the theory of [8, 9] and deduce that the F -polynomial of x coincideswith the polynomial F M associated with a certain A ℓ -module M . Futhermore M is rigid by [1, 16].By Remark 4.1 of [30], M is the kernel of a generic element of the homomorphism space betweentwo injective A ℓ -modules corresponding to the negative and positive components of the g -vector of GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 15 x . More precisely, let I ℓi,m be the injective A ℓ -module at vertex ( i, m ), then M is the kernel of ageneric element of Hom( I ℓ ( m ) − , I ℓ ( m ) + ), where I ℓ ( m ) + = M g i,s ( m ) > I ℓi,s − d i ⊕ g i,s ( m ) , I ℓ ( m ) − = M g i,s ( m ) < I ℓi,s − d i ⊕| g i,s ( m ) | . It was shown in [20] that our A ℓ -module M does not change when ℓ increases and that in thedirect limit A = lim ℓ →−∞ A ℓ . The A -module M is the kernel of a generic element of Hom( I ( m ) − , I ( m ) + ). Thus M = K ( m ). (cid:3) From the proof of Theorem 3.2 we obtain the following corollay.
Corollary 3.3.
Let L ( m ) be a prime snake module in C − . Then the generic kernel K ( m ) is rigidand indecomposable. Remark 3.4.
By Theorem 2.5, every snake module of type A n or type B n is isomorphic to a tensorproduct of prime snake modules defined uniquely up to permutation. On the other hand, if M and N are two finite-dimensional A -modules, then by Proposition 3.2 of [9] we have F M ⊕ N = F M F N .Therefore, replacing the module K ( m ) in Theorem 3.2 by a direct sum, we obtain a similar geometriccharacter formula for arbitrary snake module of types A n and B n . We present several examples to illustrate Theorem 3.2.
Example 3.5.
In type A , let N = 3 , k = 1 , k = 2 , k = 1 , i = 1 , i = 3 , i = 2 , r = − , j = j = 0 , n = 4 , and n = 5 . Then m = Y , − Y , − Y , − Y , − . We get g i,s ( m ) = if ( i, s ) = (1 , − , (3 , − , or (2 , − − if ( i, s ) = (1 , − , (3 , − , or (2 , − otherwise . Thus by Definition 2.7 I ( m ) + = I , − ⊕ I , − ⊕ I , − , I ( m ) − = I , − ⊕ I , − ⊕ I , − . The module K ( m ) has dimension 13 and is displayed in Figure 6. In Figure 6, all vertices carrya vector space of dimension . Applying Theorem 3.2, we can compute its q -character as follows.There are 160 submodules in K ( m ) . χ − q ( L ( m )) = m ((1 + v , − + v , − v , − )( v , − + v , − v , − + v , − v , − v , − + v , − v , − v , − v , − + v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − v , − v , − v , − )+ v , − v , − v , − v , − v , − ( v , − v , − + v , − v , − v , − + v , − v , − v , − v , − + v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − v , − )+ (1 + v , − + v , − v , − + v , − v , − v , − v , − v , − )(1 + v , − + v , − v , − + v , − v , − + v , − v , − v , − + v , − v , − v , − v , − )+ (1 + v , − + v , − v , − )(( v , − + v , − v , − )( v , − + v , − v , − + v , − v , − + v , − v , − v , − + v , − v , − v , − v , − )+ ( v , − v , − v , − + v , − v , − v , − v , − + v , − v , − v , − v , − v , − )( v , − + v , − + v , − v , − + v , − v , − v , − )+ ( v , − + v , − v , − )( v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − v , − v , − v , − ))+ v , − v , − v , − v , − v , − ( v , − v , − + v , − v , − v , − )( v , − + v , − + v , − v , − + v , − v , − v , − )+ v , − v , − v , − v , − v , − ( v , − + v , − v , − )( v , − v , − v , − v , − + v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − v , − )) . Starting from the initial seed ( z , G − ) , the following sequence of mutations produces (in the laststep) the cluster variable corresponding to L ( m ) . (3 , − , (2 , − , (1 , − , (2 , − , (1 , − , (2 , − , (3 , − , (1 , − , (2 , − , (3 , − , (3 , − , (2 , − , (1 , − . Example 3.6.
In type A , let N = 2 , k = 1 , k = 1 , i = 2 , i = 2 , r = − , j = 1 , and n = 4 .Then m = Y , − Y , − . We get g i,s ( m ) = if ( i, s ) = (2 , − , or (2 , − − if ( i, s ) = (2 , − , or (2 , − otherwise . Thus by Definition 2.7 I ( m ) + = I , − ⊕ I , − , I ( m ) − = I , − ⊕ I , − . GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 17 (1 , − , − , − , −
14) (2 , − , − , − , − , −
13) (3 , − , − , − , − O O z z ttttt $ $ ❏❏❏❏❏ O O $ $ ❏❏❏❏❏ $ $ ❏❏❏❏❏ z z ttttt $ $ ❏❏❏❏❏ O O z z ttttt O O $ $ ❏❏❏❏❏ $ $ ❏❏❏❏❏ z z ttttt O O O O Figure 6.
The support of the A -module K ( Y , − Y , − Y , − Y , − ) in type A . The module K ( m ) has dimension 8 and is displayed in Figure 7. In Figure 7, all vertices carrya vector space of dimension . Applying Theorem 3.2, we can compute its q -character as follows.There are 35 submodules in K ( m ) . χ − q ( L ( m )) = m (1 + v , − + v , − v , − + v , − v , − + v , − v , − v , − + v , − v , − v , − v , − v , − + ( v , − + v , − v , − + v , − v , − + v , − v , − v , − + v , − v , − v , − v , − )(1 + v , − + v , − v , − + v , − v , − + v , − v , − v , − )+ v , − v , − v , − v , − v , − ( v , − + v , − + v , − v , − + v , − v , − v , − )) . Starting from the initial seed ( z , G − ) , the following sequence of mutations produces (in the laststep) the cluster variable corresponding to L ( m ) . (3 , − , (2 , − , (1 , − , (2 , − , (3 , − , (2 , − , (1 , − , (2 , − . (1 , − , −
8) (2 , − , − , − , −
9) (3 , − , − z z ttttt $ $ ❏❏❏❏❏ $ $ ❏❏❏❏❏ $ $ ❏❏❏❏❏ z z ttttt z z ttttt O O z z ttttt $ $ ❏❏❏❏❏ Figure 7.
The support of the A -module K ( Y , − Y , − ) in type A . Example 3.7.
In type B , let N = 2 , k = 1 , k = 1 , i = 2 , i = 2 , r = − , j = 1 , and n = 6 .Then m = Y , − Y , − . We get g i,s ( m ) = if ( i, s ) = (2 , − , or (2 , − − if ( i, s ) = (2 , − , or (2 , − otherwise . Thus by Definition 2.7 I ( m ) + = I , − ⊕ I , − , I ( m ) − = I , − ⊕ I , − . The module K ( m ) has dimension 6 and is displayed in Figure 8. In Figure 8, all vertices carrya vector space of dimension . Applying Theorem 3.2, we can compute its q -character as follows.There are 15 submodules in K ( m ) . χ − q ( L ( m )) = m (1 + v , − + v , − v , − + v , − v , − v , − v , − + (1 + v , − + v , − v , − )( v , − + v , − v , − + v , − v , − v , − )+ v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − v , − ) . Starting from the initial seed ( z , G − ) , the following sequence of mutations produces (in the laststep) the cluster variable corresponding to L ( m ) . (2 , , (1 , − , (2 , − , (2 , − , (1 , − , (2 , − . GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 19 (2 , −
1) (1 , − , − , − , −
9) (2 , − O O z z ttttt $ $ ❏❏❏❏❏ $ $ ❏❏❏❏❏ z z ttttt Figure 8.
The support of the A -module K ( Y , − Y , − ) in type B .3.2. Combinatorial character formula.
Recall from Section 2.3 that for ( i, k ) ∈ G − , we denoteby p + i,k (respectively, p − i,k ) the unique highest (respectively, lowest) path in P i,k . For an arbitrarypath p ∈ P i,k , we let p ⊖ p + i,k denote the symmetric difference between p and p + i,k , defined as p ⊖ p + i,k = ( p ∪ p + i,k ) \ ( p ∩ p + i,k ). Then the set p ⊖ p + i,k encloses the union of some consecutive cells. Remark 3.8.
In type A n , the mapping p p ⊖ p + i,k defines a bijection between P i,k and the setof all Young diagrams inside an i × ( n + 1 − i ) rectangle. We define the height monomial h ( p ) of a path p ∈ P i,k by h ( p ) = Y ( i,r ) ∈ p ⊖ p + i,k v i,r , where ( i, r ) ∈ Γ − runs over all the cell coordinates in p ⊖ p + i,k and we use the convention: v i,r = 0if ( i, r ) Γ − . In particular, h ( p + i,k ) = 1.Recall that for any snake ( i t , k t ), 1 ≤ t ≤ T ∈ Z ≥ , P ( i t ,k t ) ≤ t ≤ T = { ( p , . . . , p T ) : p t ∈ P i t ,k t , ≤ t ≤ T, ( p , . . . , p T ) is non-overlapping } . Theorem 3.9.
Let L ( m ) = L ( Q Ti =1 Y i t ,k t ) be a prime snake module and K ( m ) be the generic kernelassociated to L ( m ) . Then F K ( m ) = X ( p ,...,p T ) ∈ P ( it,kt )1 ≤ t ≤ T T Y t =1 h ( p t ) . Proof.
By Theorem 3.2, we have F K ( m ) = χ − q ( L ( m )) m , (3.2) and Theorem 2.3 gives a formula for χ q ( L ( m )) in terms of paths χ q ( L ( m )) = X ( p ,...,p T ) ∈ P ( iℓ,kℓ )1 ≤ ℓ ≤ T T Y ℓ =1 m ( p ℓ ) . (3.3)Note that equation (3.2) uses the truncated q -character χ − q ( L ( m )) whereas equation (3.3) usesthe complete q -character χ q ( L ( m )). First we prove the statement in the case where χ − q ( L ( m )) = χ q ( L ( m )).Applying Lemma 5.10 of [24] and using induction, we have T Y t =1 m ( p t ) = T Y t =1 m ( p + i t ,k t ) R Y r =1 A − j r ,ℓ r . (3.4)where ( j r , ℓ r ), 1 ≤ r ≤ R ∈ Z ≥ , is a sequence of cell coordinates determined by the symmetricdifference p + i t ,k t ⊖ p t , 1 ≤ t ≤ T . Therefore T Y t =1 h ( p t ) = R Y r =1 A − j r ,ℓ r . (3.5)Moreover, since p + i t ,k t is the highest path in P i t ,k t , Theorem 2.3 implies that Q Tt =1 m ( p + i t ,k t ) containsno negative powers. Since L ( m ) is special, its highest weight monomial m is the unique dominantmonomial in χ q ( L ( m )), and thus m = T Y t =1 m ( p + i t ,k t ) . (3.6)Thus equations (3.2)–(3.6) imply mF K ( m ) = χ q ( L ( m )) = X ( p ,...,p T ) ∈ P ( iℓ,kℓ )1 ≤ ℓ ≤ T m R Y r =1 A − j r ,ℓ r = X ( p ,...,p T ) ∈ P ( it,kt )1 ≤ t ≤ T m T Y t =1 h ( p t ) . Now suppose χ − q ( L ( m )) = χ q ( L ( m )). Then we have to modify the above argument as follows.Equation (3.3) is replaced by χ − q ( L ( m )) = X ( p ,...,p T ) ∈ P ( iℓ,kℓ )1 ≤ ℓ ≤ T C + pℓ ,C − pℓ ⊂ G − T Y ℓ =1 m ( p ℓ ) . (3.7)In other words, we require that for each path p ℓ the upper and lower corners C + p ℓ , C − p ℓ lie in G − .Moreover, in equation (3.5), we replace A − j r ,ℓ r by A ′− j r ,ℓ r where A ′− j r ,ℓ r = ( A − j r ,ℓ r if ( j r , ℓ r ) ∈ Γ − , . GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 21 Then mF K ( m ) = χ − q ( L ( m )) = X ( p ,...,p T ) ∈ P ( iℓ,kℓ )1 ≤ ℓ ≤ T C + pℓ ,C − pℓ ⊂ G − m R ′ Y r =1 A − j r ,ℓ r = X ( p ,...,p T ) ∈ P ( it,kt )1 ≤ t ≤ T m R Y r =1 A ′− j r ,ℓ r = X ( p ,...,p T ) ∈ P ( it,kt )1 ≤ t ≤ T m T Y t =1 h ( p t ) . (cid:3) Remark 3.10. (1)
Theorem 3.9 allows us to calculate the dimension vector ( d i,r ( K ( m ))) ( i,r ) ∈ Γ − of the A -module K ( m ) in a combinatorial way using all T -tuples of non-overlapping paths.We will explain this in the next section. (2) Theorem 3.9 provides a combinatorial approach to find all submodules of K ( m ) , see Exam-ples 3.14-3.16. (3) Using Proposition 3.2 of [9] , for any two finite-dimensional A -modules M and N , we have F M ⊕ N = F M F N . Replacing the A -module K ( m ) in Theorem 3.9 by a direct sum of such modules, we obtaina similar combinatorial formula for arbitrary snake modules. Corollary 3.11. If L ( m ) = L ( Q Ti =1 Y i t ,k t ) is a snake module and K ( m ) is the associated generickernel, then for all dimension verctors e we have χ ( Gr e ( K ( m ))) = 0 or . Proof.
Using Theorem 3.9 and the definition of the F -polynomial, it suffices to show that for any two T -tuples ( p , . . . , p T ) = ( p ′ , . . . , p ′ T ) ∈ P ( i t ,k t ) ≤ t ≤ T of non-overlapping paths, we have Q Tt =1 h ( p t ) = Q Tt =1 h ( p ′ t ). This holds because ( p , . . . , p T ) are disjoint paths and each p i is determined by p + i ⊖ p i . (cid:3) Remark 3.12.
Corollary 3.11 holds for any thin and real module if the Conjecture 13.2 of [18] orConjecture 5.2 of [19] or Conjecture 9.1 of [22] holds.
Generic kernel.
Recall that P ( i,k ) is a collection of paths defined in Section 2.3. Let P ′ ( i,k ) = { p ∈ P ( i,k ) | C ± p ⊂ G − } ⊂ P ( i,k ) . Let P ( i t ,k t ) ≤ t ≤ T be a collection of paths associated to a snake module L ( m ) of the form (2.4)in C − . For any snake ( i t , k t ), 1 ≤ t ≤ T ∈ Z ≥ , let P ′ ( i t ,k t ) ≤ t ≤ T = { ( p ′ , . . . , p ′ T ) : p ′ t ∈ P ′ i t ,k t , ≤ t ≤ T, ( p ′ , . . . , p ′ T ) is non-overlapping } . Let V m be the set of all the cell coordinates in the set S ≤ t ≤ T ( p ′− i t ,k t ⊖ p + i t ,k t ), where p ′− i t ,k t is aminimal path in P ′ ( i t ,k t ) for each 1 ≤ t ≤ T and ( p ′− i ,k , . . . , p ′− i T ,k T ) ∈ P ′ ( i t ,k t ) ≤ t ≤ T . Note that when χ − q ( L ( Q Ti =1 Y i t ,k t )) = χ q ( L ( Q Ti =1 Y i t ,k t )), the set V m is the set of all the cellcoordinates in the set S ≤ t ≤ T ( p − i t ,k t ⊖ p + i t ,k t ). Definition 3.13.
Let Q ( m ) be the full subquiver of Γ − with vertex set V m . If we assign a vector space whose dimension is equal to the multiplicity of cells with coordinate( i, r ) occurring in the multiset S ≤ t ≤ T ( p ′− i t ,k t ⊖ p + i t ,k t ) to every point ( i, r ) ∈ V m , then by Theo-rem 3.9, the generic kernel K ( m ) is a representation of Q ( m ). In general K ( m ) is not unique, noteven up to isomorphism, but its F -polynomial is unique. In particular, the linear maps associatedwith arrows satisfy relations in the Jacobian ideal J .The following several examples hold that χ − q ( L ( Q Ti =1 Y i t ,k t )) = χ q ( L ( Q Ti =1 Y i t ,k t )). Example 3.14.
In type A , let m = Y , − Y , − Y , − Y , − . Then K ( m ) is displayed in Figure 9(Here K ( m ) is drawn opposite as Figure 6, because of the definition of paths). For each vertex ( i, r ) ∈ V m , we find it convenient to always label the dimension of the vector space at the vertex ( i, r ) . The dimension associated with a vertex ( i, r ) ∈ V m is the multiplicity of cells with coordinate ( i, r ) occurring in the multiset ( p − , − ⊖ p +1 , − ) ∪ ( p − , − ⊖ p +3 , − ) ∪ ( p − , − ⊖ p +3 , − ) ∪ ( p − , − ⊖ p +2 , − ) . The maps associated with arrows are ( ± , whose sign is deduced from the defining relations of theJacobian algebra A .In the sense of Theorem 3.9, finding all possible submodules of K ( m ) is equivalent to finding all4-tuple sets of non-overlapping paths in P (1 , − × P (3 , − × P (3 , − × P (2 , − . Example 3.15.
In type A , let m = Y , − Y , − . Then K ( m ) is displayed in Figure 10 (Here K ( m ) is drawn opposite as Figure 7, because of the definition of paths). For each vertex ( i, r ) ∈ V m , welabel the dimension of the vector space at the vertex ( i, r ) . The dimension associated with a vertex ( i, r ) ∈ V m is the multiplicity of cells with coordinate ( i, r ) occurring in the multiset ( p − , − ⊖ p +2 , − ) ∪ ( p − , − ⊖ p +2 , − ) . The maps associated with arrows are ( ± , whose sign is deduced from the defining relations of theJacobian algebra A .In the sense of Theorem 3.9, finding all possible submodules of K ( m ) is equivalent to finding allpairs of non-overlapping paths in P (2 , − × P (2 , − . Example 3.16.
In type B , let m = Y , − Y , − . Then K ( m ) is displayed in Figure 11 (Here K ( m ) is drawn opposite as Figure 8, because of the definition of paths). For each vertex ( i, r ) ∈ V m , welabel the dimension of the vector space at the vertex ( i, r ) . The dimension associated with a vertex ( i, r ) ∈ V m is the multiplicity of cells with coordinate ( i, r ) occurring in the multiset ( p − , − ⊖ p +2 , − ) ∪ ( p − , − ⊖ p +2 , − ) . The maps associated with arrows are ( ± , whose sign is deduced from the defining relations of theJacobian algebra A .In the sense of Theorem 3.9, finding all possible submodules of K ( m ) is equivalent to finding allpairs of non-overlapping paths in P (2 , − × P (2 , − . The following is an example where the dimensions of K ( m ) are larger than 1. Example 3.17.
In type A , let m = Y , − Y , − . Then L ( m ) is a Kirillov-Reshetikhin module and K ( m ) is displayed in Figure 12. For each vertex ( i, r ) ∈ V m , we label the dimension of the vector GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 23 − − − − − − − − − − − − − − • • •••• ••• ••• • Figure 9.
The module K ( Y , − Y , − Y , − Y , − ) in type A . The index of Y i,r corresponds to the top vertex of the associated path rectangle P i,r . The dimensionis 1 at those vertices of Γ − that lie in the interior of these rectangles. space at the vertex ( i, r ) . The dimension associated with a vertex ( i, r ) ∈ V m is the multiplicity ofcells with coordinate ( i, r ) occurring in the multiset ( p − , − ⊖ p +2 , − ) ∪ ( p − , − ⊖ p +2 , − ) . In Figure 12,almost all vertices carry a vector space of dimension 1, except the vertex (2 , − which carries avector space of dimension 2.Starting from the initial seed ( z , G − ) , the following sequence of mutations produces (in the laststep) the cluster variable corresponding to L ( m ) . (2 , − , (2 , − , (2 , − , (1 , − , (1 , − , (3 , − , (3 , − , (2 , − , (2 , − . − − − − − − − − − •••••• •• Figure 10.
The module K ( Y , − Y , − ) in type A . The index of Y i,r correspondsto the top vertex of the associated path rectangle P i,r . The dimension is 1 at thosevertices of Γ − that lie in the interior of these rectangles.1 2 10 − − − − − − • ••• ••• • Figure 11.
The module K ( Y , − Y , − ) in type B . The index of Y i,r correspondsto the top vertex of the associated path triangle P i,r . The dimension is 1 at thosevertices of Γ − that lie in the interior of these triangles. GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 25 − − − − − − − •••••• •• Figure 12.
The module K ( Y , − Y , − ) in type A . The index of Y i,r correspondsto the top vertex of the associated path rectangle P i,r . The dimension is 1 at thosevertices of Γ − that lie in the interior of these rectangles, except that the vertex(2 , −
5) has a vector space of dimension 2.
Remark 3.18.
The dimension of K ( m ) at a vertex ( i, r ) can be arbitrary large in the sense thatgiven any integer α there is a snake module L ( m ) and a vertex ( i, r ) such that the generic kernel K ( m ) is of dimension at least α at ( i, r ) . Therefore Corollary 3.11 is non-trival. The following is an example that χ − q ( L ( Q Ti =1 Y i t ,k t )) = χ q ( L ( Q Ti =1 Y i t ,k t )). Example 3.19.
In type A , let m = Y , − Y , − . Then L ( m ) is a Kirillov-Reshetikhin module and K ( m ) is displayed in Figure 13. By definition, we have V m = ( p ′− , − ⊖ p +2 , − ) ∪ ( p ′− , − ⊖ p +2 , − ) = { (2 , − , (2 , − } , where p ′− , − = { (0 , , (1 , − , (2 , , (3 , − , (4 , } and p ′− , − = { (0 , − , (1 , − , (2 , − , (3 , − , (4 , − } . Note that p ′− , − is a path in the set P ′ , − , so it cannot go through points ( i, r ) with r > .For each vertex ( i, r ) ∈ V m , we label the dimension of the vector space at the vertex ( i, r ) . Thedimension associated with a vertex ( i, r ) ∈ V m is the multiplicity of cells with coordinate ( i, r ) occurring in the multiset ( p ′− , − ⊖ p +2 , − ) ∪ ( p ′− , − ⊖ p +2 , − ) . In Figure 13, all vertices carry a vectorspace of dimension 1.Starting from the initial seed ( z , G − ) , the sequence ((2 , , (2 , − of mutations produces (in thelast step) the cluster variable corresponding to L ( m ) . Denominator vector
In this section, we show that every snake module corresponds to a cluster monomial with squarefree denominator in the cluster algebra A and that snake modules are real modules. − − − − •• Figure 13.
The module K ( Y , − Y , − ) in type A . The index of Y i,r correspondsto the top vertex of the associated path rectangle P i,r . The dimension is 1 atvertices (2 , −
3) and (2 , − Theorem 4.1.
Let L ( m ) be an arbitrary snake module. Then the truncated q -character χ − q ( L ( m )) is a cluster monomial in A , and its denominator is square free as a monomial in the initial clustervariables z i,r , ( i, r ) ∈ G − .Proof. By Theorem 2.5, we can write L ( m ) as a tensor product L ( m ) ∼ = L ( m ) ⊗ · · · ⊗ L ( m n ) ofprime snake modules. Let K ( m i ) be the generic kernel associated to L ( m i ) and let Q ( m i ) be thefull subquiver of Γ − whose vertices are in the support of K ( m i ). Thus K ( m ) = L ni =1 K ( m i ) is thegeneric kernel associated to L ( m ). To show that L ( m ) corresponds to a cluster monomial we needto prove that K ( m ) is a rigid object in the cluster category [1, 16].Let P ( i t ( m i ) ,k t ( m i )) ≤ t ≤ Ti be the set of paths associated to L ( m i ). By Theorem 2.3 and Re-mark 2.4, we know that for any 1 ≤ i = j ≤ n , the sets P ( i t ( m i ) ,k t ( m i )) ≤ t ≤ Ti and P ( i t ( m j ) ,k t ( m j )) ≤ t ≤ Tj are non-overlapping. By our construction in Section 3.3, this implies that the quivers Q ( m i ) and Q ( m j ) are disjoint and there are no arrows in Γ − which connect Q ( m i ) and Q ( m j ).Definition 2.7 and Proposition 3.1 imply that for the prime snake module L ( m j ), we have I ( m j ) + = M g ℓ,s ( m j )=1 I ℓ,s − d ℓ , I ( m j ) − = M g ℓ,s ( m j )= − I ℓ,s − d ℓ . By Section 3.3, the support of K ( m i ) is contained in P ( i t ( m i ) ,k t ( m i )) ≤ t ≤ Ti . Thus the socle points( ℓ, s − d ℓ ) in I ( m j ) + cannot be in the support of K ( m i ). Otherwise, the set P ℓ,s for L ( m j ) and P ( i t ( m i ) ,k t ( m i )) ≤ t ≤ Ti would be overlapping (there is at least a common vertex ( ℓ, s )). This is acontradiction to the fact that L ( m i ) ⊗ L ( m j ) is not prime, see Remark 2.4. GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 27 Therefore Hom A ( K ( m i ) , I ( m j ) + ) = Hom A ( K ( m i ) , M g ℓ,s ( m j )=1 I ℓ,s − d ℓ ) ∼ = M g ℓ,s ( m j )=1 ( K ( m i )) ℓ,s − d ℓ = 0 . (4.1)Similarly, Hom A ( K ( m j ) , I ( m i ) + ) = 0.Consider the injective resolution0 / / K ( m j ) i / / I − ( m j ) i / / I ( m j ) + i / / · · · . Then Ext A ( K ( m i ) , K ( m j )) is a quotient of { f ∈ Hom( K ( m i ) , I ( m j ) + ) | i f = 0 } which is zeroby (4.1). Thus Ext A ( K ( m i ) , K ( m j )) = 0 . Similarly, Ext A ( K ( m j ) , K ( m i )) = 0.By Corollary 3.3, we have that Ext A ( K ( m i ) , K ( m i )) = 0 for any 1 ≤ i ≤ n . In [1], K ( m i ) and K ( m j ) are compatible if and only ifExt C ( K ( m i ) , K ( m j )) = 0 , where C is the (generalized) cluster category of the Jacobian algebra A .Applying Ext C ( K ( m i ) , K ( m j )) ∼ = Ext A ( K ( m i ) , K ( m j )) M Ext A ( K ( m j ) , K ( m i )) , we see that K ( m i ) and K ( m j ) are compatible for all i, j = 1 , . . . , n , and hence snake modules arecluster monomials.Next we prove the statement about square free denominators using the Mukhin-Young’s formulasin Theorem 2.3. By Theorem 3.9 and its proof, we have χ − q ( L ( m )) = mF K ( m ) = X ( p ,...,p T ) ∈ P ( it,kt )1 ≤ t ≤ T m T Y t =1 h ( p t ) . For any T -tuple ( p , . . . , p T ) of non-overlapping paths, either m Q Tt =1 h ( p t ) = 0 or by Theorem 2.3, m T Y t =1 h ( p t ) = T Y t =1 m ( p t ) = T Y t =1 Y ( j,ℓ ) ∈ C + pt Y j,ℓ Y ( j,ℓ ) ∈ C − pt Y − j,ℓ = T Y t =1 Y ( j,ℓ ) ∈ C + pt z j,ℓ z j,ℓ + b jj Y ( j,ℓ ) ∈ C − pt z j,ℓ + b jj z j,ℓ , (4.2)where the last equation is obtained by performing the change of variables (2.1). It is obvious that Y ( j,ℓ ) ∈ C + pt z j,ℓ z j,ℓ + b jj Y ( j,ℓ ) ∈ C − pt z j,ℓ + b jj z j,ℓ is a fraction with square free denominator in the initial cluster variables z i,r , ( i, r ) ∈ G − . For any 1 ≤ t = t ≤ T , the expression Y ( j,ℓ ) ∈ C + pt z j,ℓ z j,ℓ + b jj Y ( j,ℓ ) ∈ C − pt z j,ℓ + b jj z j,ℓ Y ( j,ℓ ) ∈ C + pt z j,ℓ z j,ℓ + b jj Y ( j,ℓ ) ∈ C − pt z j,ℓ + b jj z j,ℓ (4.3)is still a fraction with square free denominator. Otherwise either z j,ℓ + b jj for some ( j, ℓ ) ∈ C + p t or z j,ℓ for some ( j, ℓ ) ∈ C − p t in the first term also appear in the denominator of the second term. If( j, ℓ ) ∈ C + p t , then p t and p t overlap at least at the vertex ( j, ℓ ). If ( j, ℓ ) ∈ C − p t , then p t and p t overlap at least at a vertex ( i, r ), where ι − ( i, r ) = ( j ± , ℓ + 1) or ( j ± , ℓ + 2). This is acontradiction. Similarly we deal with ( j, ℓ ) ∈ C − p t .Therefore (by induction) χ − q ( L ( m )) is a Laurent polynomial with square free denominator in theinitial cluster variables. (cid:3) Recall that prime snake modules are prime, real modules, see Theorem 2.5. As a slight general-ization, we have the following theorem.
Theorem 4.2.
Snake modules are real simple modules.Proof.
We assume that L ( m ) is a snake module. Then L ( m ) = L ( m ) ⊗ · · · ⊗ L ( m n ) with L ( m i )prime by Theorem 2.5. We only need to show that snake modules are real.Using the fact that χ q is a ring homomorphism, we have χ q ( L ( m ) ⊗ L ( m )) = χ q ( L ( m )) χ q ( L ( m ))= χ q ( L ( m )) · · · χ q ( L ( m n )) χ q ( L ( m )) · · · χ q ( L ( m n ))= χ q ( L ( m ) ⊗ L ( m )) · · · χ q ( L ( m n ) ⊗ L ( m n )) . By Theorem 3.4 of [10], we have the fact that for every 1 ≤ ℓ ≤ n , χ q ( L ( m ℓ ) ⊗ L ( m ℓ )) has onlyone dominant monomial m ℓ .Using Theorem 2.3 and Remark 2.4, for any 1 ≤ i = j ≤ n , we see that P ( i t ( m i ) ,k t ( m i )) ≤ t ≤ Ti and P ( i t ( m j ) ,k t ( m j )) ≤ t ≤ Tj are non-overlapping. So monomials with negative exponents occurring in χ q ( L ( m i ) ⊗ L ( m i )) cannotbe canceled by any monomial occurring in χ q ( L ( m j ) ⊗ L ( m j )). Thus χ q ( L ( m )) χ q ( L ( m )) = χ q ( L ( m ) ⊗ L ( m ))has only one dominant monomial m . This shows that L ( m ) ⊗ L ( m ) is simple, and thus L ( m ) isreal. (cid:3) Remark 4.3. (1)
From Proposition 3.1, it follows that for different snake modules, the corre-sponding cluster monomials have different g -vectors with respect to a given initial seed. (2) Combining Theorem 4.1 and Theorem 4.2, we give a partial answer of Conjecture 1.1.
Recall that A is the cluster algebra introduced by Hernandez and Lerclec in [20], also see Sec-tion 2.1. In [20], Hernandez and Leclerc applied the method of cluster mutations to give an algorithmfor computing the q -characters of Kirillov-Reshetikhin modules by successive approximations.An explicit formula of the expansion for snake modules even Kirillov-Reshetikhin modules interms of the initial cluster variables is usually very complicated. In the following theorem, we giveexplicitly the denominator of the cluster monomial associated to a snake module L ( m ). GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 29 Theorem 4.4.
Suppose that L ( m ) is a snake module. Then the denominator of the cluster mono-mial associated to L ( m ) is d ( χ − q ( L ( m ))) = Y ( i,r − d i ) ∈ Supp ( K ( m )) z i,r . (4.4) Proof.
We first show that for any ( i, r − d i ) ∈ Supp( K ( m )), the variable z i,r appears in the denom-inator of the cluster monomial associated to L ( m ).We assume without loss of generality that m = Q Ti =1 Y i t ,k t for some snake ( i t , k t ) ≤ t ≤ T . Thecluster monomial of L ( m ) is given by the truncated q -character χ − q ( L ( m )) after the change ofvariables (2.1). For any ( i, r ) such that ( i, r − d i ) ∈ Supp( K ( m )), there exists a path p t ∈ P ′ i t ,k t such that ( i, r ) is the unique lower corner of p t . We choose p t such that t is maximal. Define p ℓ ∈ P ′ i ℓ ,k ℓ by p ℓ = ( p + i ℓ ,k ℓ ≤ ℓ < t,p ′− i ℓ ,k ℓ t < ℓ ≤ T. Then the T -tuple ( p , . . . , p T ) is a set of non-overlapping paths, because p t is a path between p + i t ,k t and p ′− i t ,k t .Using equation (3.7), we have χ − q ( L ( m )) = X ( p ,...,p T ) ∈ P ( iℓ,kℓ )1 ≤ ℓ ≤ T C + pℓ ,C − pℓ ⊂ G − T Y ℓ =1 m ( p ℓ ) . After performing the change of variables (2.1), the variable z i,r appears in the denominator of Q Tℓ =1 m ( p ℓ ). Thus the product (4.4) appears in the denominator of the cluster monomial χ − q ( L ( m )).On the other hand, for any T -tuple ( p , . . . , p T ) of non-overlapping paths, we have the following:If z i,r , ( i, r ) ∈ G − , appears in the denominator of Q Tℓ =1 m ( p ℓ ), then ( i, r ) ∈ C − p ℓ or ( i, r − b ii ) ∈ C + p ℓ for some ℓ by equation (4.2). For the lower corner ( i, r ) ∈ C − p ℓ , we have ( i, r − d i ) ∈ Supp( K ( m )),by Remark (3.10) (1). For the upper corner ( i, r − b ii ) ∈ C + p ℓ , we need to modify our path p ℓ byreplacing the point ( i, r − b ii ) by the point ( i, r ). Note that ( i, r ) ∈ G − , so the modified path is stillin P ′ ( i ℓ ,k ℓ ) . Now Remark (3.10) (1) implies ( i, r − d i ) ∈ Supp( K ( m )). (cid:3) We illustrate this result in our five running examples. Note that for each vertex ( i, r − d i ) in thesupport of K ( m ), as shown in Figures 9–13, we have a contribution z i,r in the denominator. Here d i = 1 in type A and d = 2 and d = 1 in type B . Example 4.5.
In type A , let m = Y , − Y , − Y , − Y , − . Then by Theorem 4.4, d ( χ − q ( L ( m ))) = z , − z , − z , − z , − z , − z , − z , − z , − z , − z , − z , − z , − z , − . Example 4.6.
In type A , let m = Y , − Y , − . Then by Theorem 4.4, d ( χ − q ( L ( m ))) = z , − z , − z , − z , − z , − z , − z , − z , − . Example 4.7.
In type B , let m = Y , − Y , − . Then by Theorem 4.4, d ( χ − q ( L ( m ))) = z , z , − z , − z , − z , − z , − . Example 4.8.
In type A , let m = Y , − Y , − . Then by Theorem 4.4, d ( χ − q ( L ( m ))) = z , − z , − z , − z , − z , − z , − z , − . Example 4.9.
In type A , let m = Y , − Y , − . Then by Theorem 4.4, d ( χ − q ( L ( m ))) = z , z , − . It is natural to ask whether all cluster variables with square free denominator are always primesnake modules. The answer is No. The following example shows that there exists a module that isnot a snake module and such that its truncated q -character corresponds to a cluster variable withsquare free denominator. Example 4.10.
In type A , let m = Y , − Y , Y , − . This is not a snake module. Because thesecond coordinates in the indices do not form an increasing sequence. By Example 12.2 of [18] , wehave [ L ( Y , − Y , Y , − )][ L ( Y , )] = [ L ( Y , − Y , )][ L ( Y , − Y , )] + [ L ( Y , − Y , )] . Thus by Theorem 5.1 of [20] χ − q ( L ( m )) = χ − q ( L ( Y , − Y , )) χ − q ( L ( Y , − Y , )) + χ − q ( L ( Y , − Y , )) χ − q ( L ( Y , )) . On the right hand side of the equation, every truncated q -character is known by the Frenkel-Mukhinalgorithm, so χ − q ( L ( m )) = m (1 + A − , − + A − , − + A − , − A − , − + A − , − A − , − A − , − ) . The corresponding cluster variable x m is x m = ( z , − z , − + z , z , − ) z , − + z , − + ( z , z , − + z , z , − ) z , − z , − z , z , − , which has a square free denominator. Finally, we point out that there exists a module beyond snake modules in C − for which Hernandezand Leclerc’s conjectural geometric formula holds. Example 4.11.
In type A , let m = Y , − Y , − Y , − . By [18] , we know that L ( m ) corresponds toa cluster variable in A (up to scalar), equivalently, its truncated q -character is a cluster variablein A . Let I ( m ) + = I , − ⊕ I , − ⊕ I , − , I ( m ) − = I , − ⊕ I , − ⊕ I , − . By Example 12.2 of [18] , we have [ L ( Y , − Y , − Y , − )][ L ( Y , − )] = [ L ( Y , − Y , − )][ L ( Y , − Y , − )] + [ L ( Y , − Y , − )] . With the exception of L ( Y , − Y , − Y , − ) , those modules are minimal affinizations [2] , and we cancompute their q -characters by the Frenkel-Mukhin algorithm.On the other hand, the formula in Theorem 3.2 holds for L ( Y , − Y , − Y , − ) . The module K ( m ) has dimension 10 and is displayed in Figure 14. In Figure 14, almost all vertices carry a vectorspace of dimension 1, except the vertex (2 , − which carries a vector space of dimension 2. Themaps associated with the arrows incident to (2 , − have the following matrices: α = (cid:0) (cid:1) , β = (cid:0) (cid:1) , γ = (cid:0) (cid:1) , δ = (cid:18) (cid:19) , η = (cid:18) (cid:19) . GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 31 (1 , −
2) (2 , − , −
3) (3 , − , −
6) (2 , −
5) (3 , − , − , − z z ttttt $ $ ❏❏❏❏❏ $ $ ❏❏❏❏❏ z z ttttt η z z ttttt γ z z ttttt δ $ $ ❏❏❏❏❏ β $ $ ❏❏❏❏❏ O O O O α O O Figure 14.
The A -module K ( m ) for m = Y , − Y , − Y , − in type A . All other arrows carry linear maps with ( ± , whose sign is easily deduced from the defining relationsof the Jacobian algebra A . There are 70 submodules in K ( m ) .Then χ − q ( L ( m )) = m ((1 + v , − + v , − + v , − v , − )(1 + v , − + v , − v , − + v , − v , − + v , − v , − v , − + v , − v , − v , − v , − )+ ( v , − v , − + v , − v , − )( v , − + v , − v , − + v , − v , − + v , − v , − v , − + v , − v , − v , − v , − )+ v , − v , − v , − v , − v , − (1 + v , − + v , − v , − )+ v , − v , − v , − v , − v , − (1 + v , − + v , − v , − )+ v , − v , − v , − v , − v , − v , − (1 + v , − + v , − v , − )+ v , − v , − v , − v , − v , − v , − (1 + v , − + v , − v , − )+ v , − v , − v , − + 2 v , − v , − v , − v , − + 2 v , − v , − v , − v , − v , − + 2 v , − v , − v , − v , − v , − + 2 v , − v , − v , − v , − v , − v , − + 2 v , − v , − v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − (1 + v , − + v , − + v , − v , − + v , − v , − v , − )+ v , − v , − v , − v , − v , − v , − v , − (1 + v , − + v , − v , − )+ v , − v , − v , − v , − v , − v , − v , − (1 + v , − + v , − v , − )+ v , − v , − v , − v , − v , − v , − v , − v , − v , − + v , − v , − v , − v , − v , − v , − v , − v , − v , − v , − ) , in agreement with Hernandez and Leclerc’s conjectural geometric formula and its denominator isnot square free. Indeed, there exists a submodule of K ( m ) whose support at vertices (2 , − , (1 , − , (3 , − such that mv , − v , − v , − = mA − , − A − , − A − , − = z , − z , − z , − z , z , − z , − z , − z , − z , − z , − . Moreover, the simple module L ( Y , − Y , − Y , − ) is not special, because there are two dominantmonomials m = Y , − Y , − Y , − and mv , − v , − v , − = Y , − in χ q ( L ( Y , − Y , − Y , − )) . Thesimple module L ( Y , − Y , − Y , − ) is not thin, because some terms in the q -character have coefficient > .Using the method introduced in Section 3.3, we obtain the dimension vector ( d j,s ( K ( m )) ( j,s ) ∈ N Γ − of K ( m ) as follows. d j,s ( K ( m )) = j, s ) = (2 , − , (1 , − , (3 , − , (2 , − , (1 , − , (3 , − , (1 , − , (3 , − , j, s ) = (2 , − , otherwise. Comparing Example 4.10 with Example 4.11, we reformulate the following classification question.In the cluster algebra A , are cluster variables with square free denominators in bijection withprime snake modules and some other prime real modules whose truncated q -characters are not equalto their q -characters? 5. Factorial cluster algebras
In this section, we apply the results of [11] to show that the C cluster algebra is factorial forDynkin types A , D , E .Following Section 4.2 of [18], let Q be a quiver with vertex set { , . . . , n, ′ , . . . , n ′ } subject tothe following two conditions:(1) The full subquiver on { , . . . , n } is an orientation of the associated Dynkin diagram ∆ oftype A , D or E , oriented in such a way that every vertex in I is a source and every vertexof I is a sink;(2) For every i ∈ I , one adds a frozen vertex i ′ and an arrow i ′ → i if i ∈ I and an arrow i → i ′ if i ∈ I .Obviously, the defining quiver Q is an acyclic quiver. Let A ( Q ) be the cluster algebra defined bythe initial seed ( { x , . . . , x n , y , . . . , y n } , Q ). Then A ( Q ) is the C cluster algebra of type ∆ in [18].Let x ′ , . . . , x ′ n be the n cluster variables obtained from the initial seed by one single mutation.Then, for each i , we have x i x ′ i = f i , where f i is a binomial in the initial seed. Recall from [11] thattwo vertices i, j ∈ { , , . . . , n } are called partners if f i and f j have a non-trivial common factor.Partnership is an equivalence relation and the equivalence classes are called partner sets . Theorem 5.1.
The C cluster algebra is factorial for Dynkin types A , D , E .Proof. By Corollary 5.2 of [11], we only need to show that every partner set in Q is a singleton.This holds because for every i ∈ I the exchange polynomial f i is a polynomial in the variable y i . (cid:3) We give an example to explain Theorem 5.1.
Example 5.2.
Let g be of type A . We choose I = { , } and I = { } . The quiver Q is asfollows. / / (cid:15) (cid:15) o o O O
2’ 3’ O O Here the vertices with boxes are frozen vertices and its associated exchange matrix is
GEOMETRIC q -CHARACTER FORMULA FOR SNAKE MODULES 33 − −
10 1 01 0 00 − . The exchange polynomials are f = x + y ,f = x x + y ,f = x + y . The polynomials f , f , f are pairwise coprime and hence every partner set in Q is a singleton. Remark 5.3.
In Section 7.1 of [17] , Geiss, Leclerc, and Schr¨oer proved that the cluster algebra A associated to Dynkin type A is a factorial cluster algebra. They also showed that the clustervariables in a factorial cluster algebra are prime elements. In [11, Theorem 3.10] , it was shownthat if A is a factorial cluster algebra and x is a non-initial cluster variable, then the associated F -polynomial F x is prime.It is natural to ask whether the C ℓ cluster algebras, with ℓ > , are factorial. The argument inthe proof of Theorem 5.1 does not work in this case, because we don’t know whether these clusteralgebras are of acyclic type. Acknowledgements
We would like to thank A. Garcia Elsener for explaining the results of [11] to us.
References [1] C. Amiot,
Cluster categories for algebras of global dimension 2 and quivers with potential , Ann. Inst. Fourier(Grenoble) (2009), no. 6, 2525–2590.[2] V. Chari, Minimal affinizations of representations of quantum groups: the rank case , Publ. Res. Inst. Math.Sci. (1995), no. 5, 873–911.[3] V. Chari and A. Pressley, Quantum affine algebras , Comm. Math. Phys. (1991), no. 2, 261–283.[4] ,
A guide to quantum groups , Cambridge University Press, Cambridge, 1994.[5] ,
Quantum affine algebras and their representations , Representations of groups (Banff, AB, 1994), CMSConf. Proc., vol. 16, Amer. Math. Soc., Providence, RI, 1995, pp. 59–78.[6] ,
Factorization of representations of quantum affine algebras , Modular interfaces (Riverside, CA, 1995),AMS/IP Stud. Adv. Math., vol. 4, Amer. Math. Soc., Providence, RI, 1997, pp. 33–40.[7] G. Cerulli Irelli, B. Keller, D. Labardini-Fragoso, and P. Plamondon,
Linear independence of cluster monomialsfor skew-symmetric cluster algebras , Compos. Math. (2013), no. 10, 1753–1764.[8] H. Derksen, J. Weyman, and A. Zelevinsky,
Quivers with potentials and their representations. I. Mutations ,Selecta Math. (N.S.) (2008), no. 1, 59–119.[9] , Quivers with potentials and their representations II: applications to cluster algebras , J. Amer. Math.Soc. (2010), no. 3, 749–790.[10] B. Duan, J.-R. Li, and Y.-F Luo, Cluster algebras and snake modules , J. Algebra (2019), 325–377.[11] G. Elsener, P. Lamp, and D. Smertnig,
Factoriality and class groups of cluster algebras , arXiv:1712.06512, 1–32.[12] E. Frenkel and E. Mukhin,
Combinatorics of q -characters of finite-dimensional representations of quantumaffine algebras , Comm. Math. Phys. (2001), no. 1, 23–57. [13] E. Frenkel and N. Reshetikhin, The q -characters of representations of quantum affine algebras and deformationsof W -algebras , Recent developments in quantum affine algebras and related topics (Raleigh, NC, 1998), Contemp.Math., vol. 248, Amer. Math. Soc., Providence, RI, 1999, pp. 163–205.[14] S. Fomin and A. Zelevinsky, Cluster algebras I: Foundations , J. Amer. Math. Soc. (2002), no. 2, 497–529.[15] , Cluster algebras. IV. Coefficients , Compos. Math. (2007), no. 1, 112–164.[16] C. Fu and B. Keller,
On cluster algebras with coefficients and 2-Calabi-Yau categories , Trans. Amer. Math. Soc. (2010), no. 2, 859–895.[17] C. Geiss, B. Leclerc, and J. Schr¨oer,
Factorial cluster algebras , Doc. Math. (2013), 249–274.[18] D. Hernandez and B. Leclerc, Cluster algebras and quantum affine algebras , Duke Math. J. (2010), no. 2,265–341.[19] ,
Monoidal categorifications of cluster algebras of type A and D , Symmetries, integrable systems andrepresentations, Springer Proc. Math. Stat., vol. 40, Springer, Heidelberg, 2013, pp. 175–193.[20] , A cluster algebra approach to q -characters of Kirillov-Reshetikhin modules , J. Eur. Math. Soc. (JEMS) (2016), no. 5, 1113–1159.[21] B. Leclerc, Imaginary vectors in the dual canonical basis of U q ( n ), Transform Groups (2003), no. 1, 95–104.[22] , Cluster algebras and representation theory , Proceedings of the International Congress of Mathemati-cians, Hindustan Book Agency, New Delhi, vol. IV, 2010, pp. 2471–2488.[23] K. Lee and R. Schiffler,
Positivity for cluster algebras , Ann. of Math. (2) (2015), no. 1, 73–125.[24] E. Mukhin and C. A. S. Young,
Path description of type
B q -characters , Adv. Math. (2012), no. 2, 1119–1150.[25] ,
Extended T -systems , Selecta Math. (N.S.) (2012), no. 3, 591–631.[26] H. Nakajima, Quiver varieties and t -analogs of q -characters of quantum affine algebras , Ann. of Math. (2) (2004), no. 3, 1057–1097.[27] , Quiver varieties and cluster algebras , Kyoto J. Math. (2011), no. 1, 71–126.[28] Y. Palu, Cluster characters for 2-Calabi-Yau triangulated categories , Ann. Inst. Fourier (Grenoble) (2008),no. 6, 2221–2248.[29] , Cluster characters II: a multiplication formula , Proc. Lond. Math. Soc. (3) (2012), no. 1, 57–78.[30] P.-G. Plamondon,
Generic bases for cluster algebras from the cluster category , Int. Math. Res. Not. IMRN , no. 10, 2368–2420.[31] F. Qin,
Triangular bases in quantum cluster algebras and monoidal categorification conjectures , Duke Math. J. (2017), no. 12, 2337–2442.[32] Q.-Q. Zhang, B. Duan, J.-R. Li, and Y.-F. Luo,
M-systems and cluster algebras , Int. Math. Res. Not. IMRN , no. 14, 4449–4486.
School of Mathematics and Statistics, Lanzhou University, Lanzhou 730000, P. R. China.
E-mail address : [email protected] Department of Mathematics, University of Connecticut, Storrs, CT 06269-3009, USA
E-mail address ::